Assessment of Convolutional Neural Networks for Automated Classification of Chest Radiographs

Abstract

hest radiography represents the initial imaging test for important thoracic abnormalities ranging from pneumonia to lung cancer. Unfortunately, as the ratio of image volume to qualified radiologists has continued to increase, interpretation delays and backlogs have demonstrably reduced the quality of care in large health organizations, such as the U.K. National Health Service (1) and the U.S. Department of Veterans Affairs (2). The situation is even worse in resource-poor areas, where radiology services are extremely scarce (3,4). In this light, automated image analysis represents an appealing mechanism to improve throughput while maintaining, and potentially improving, quality of care. The remarkable success of machine learning techniques such as convolutional neural networks (CNNs) for image classification tasks makes these algorithms a natural choice for automated radiograph analysis (5,6), and they have already performed well for tasks such as skeletal bone age assessment (7-9), lung nodule classification (10), tuberculosis detection (11), high-throughput image retrieval (12,13), and evaluation of endotracheal tube positioning (14). However, a major challenge when applying such techniques to chest radiography at scale has been the limited availability of the large labeled data sets generally required to achieve high levels of performance (6). In response, the U.S. National Institutes of Health released a public chest radiograph database containing 112 120 frontal view images with noisy multiclass labels extracted from associated text reports (15). This study also showed the challenges of achieving reliable multiclass thoracic diagnosis prediction with chest radiographs (15), potentially limiting the clinical utility of resultant classifiers. Further, this method of disease-specific computer-assisted diagnosis may not ultimately be beneficial to the interpreting clinician (16).

PENet—a scalable deep-learning model for automated diagnosis of pulmonary embolism using volumetric CT imaging

“…8,9 However, providing rapid and accurate diagnostic imaging is increasingly difficult to sustain for many medical systems and radiology providers as utilization has expanded; for example, CTPA usage alone in the emergency setting has increased 27-fold over the past 2 decades. 10,11 Applications of deep learning have already shown significant promise in medical imaging including chest and extremity Xrays, [12][13][14][15] head CT, 16 and musculoskeletal magnetic resonance imaging (MRI). 17 But despite the potential clinical and engineering advantages for utilization of deep learning automated PE classification on CTPA studies, significant development challenges remain when compared to other applications.…”

Cross-Modal Data Programming Enables Rapid Medical Machine Learning

“…Automated Triage of Frontal Chest Radiographs As the demand for imaging services increases, automated triage for common diagnostics such as chest radiographs is expected to become an increasingly important part of radiological workflows. 5,7 Our frontal chest radiography dataset comprises a 50,000-sample subset of the automated triage dataset described in Dunnmon et al 7 containing paired images and text reports, where each example describes a unique patient. Each image is also associated with a prospective normal or abnormal label provided by a single radiologist at the time of interpretation, and the dataset balance is 80% abnormal.…”

“…For both fully and weakly supervised models, we use a development set size of 200 prospectively labeled images for cross-validation during the training process and evaluate on the same 533-image held-out test dataset as Dunnmon et al, which is labeled by blinded consensus of two radiologists with 5 and 20 years of training, for the sake of consistency with the published literature. 7 Examples of chest radiographs can be found in Fig. 2a.…”

Computer‐aided diagnosis in the era of deep learning

“…Researchers have developed methods to visualize the feature maps at each convolutional layer inside the deep learning structure and to highlight the target objects recognized by the DCNN with a class activation map . Initial efforts have been made to use these tools to visualize the detected location of abnormalities or to visualize the characteristics of the deep features on medical images. These efforts are the first steps toward understanding the inner‐workings of deep learning but they are still far from being able to present the network response to clinicians with more insightful medical interpretation, especially for more complex applications than detection.…”